Quin Bingham

INCORPORATION OF THERMAL IMAGERY IN PARAMETER ESTIMATION AND MODEL CALIBRATION

Prepared by:

Quinten Glen Bingham

CEE 6440

GIS in Water Resource Engineering

Fall 2008

David Tarboton

Utah State University

David Maidment

University of Texas at Austin

Ayse Irmak
University of Nebraska

1. INTRODUCTION

Water temperature is one of the most influential physical properties of freshwater systems. Chemical and biological characteristics of freshwater systems are affected by water temperature (Benson and Krause 1980; Stoneman and Jones 2000). If instream temperature is permanently altered, this may render formerly suitable habitat unsuitable for native species assemblages (Holtby 1988; Wissmar, Smith et al. 1994a; Quigley and Arbelbide 1997). Due to the semiarid to arid climate of southern Utah, low stream flows are commonplace. Therefore, heightened demands for water in Washington County Utah, primarily due to population increases, have directly impacted already low flows, thus affecting the instream temperature regime of the Virgin River. Two endangered species that are unique to the Virgin River currently exist and have historically suffered from elevated instream temperatures; the Virgin River Chub (Gila seminuda) and the Woundfin (Plagopterus argentissimus). Therefore, these endangered species are of chief management concern. In 1972, The Clean Water act was created and instituted, which requires that water quality standards for temperature are established and met, based on the needs of sensitive species. Once these standards are set, states must understand when these limiting conditions occur, what created the impairent, and which management options will remedy the impairment. Ninth on the list of the Top 100 U.S. Environmental Protection Agency (EPA) water quality impairments is temperature impaired water systems (EPA 2004).

In efforts to mitigate elevated instream temperatures of the Virgin River, research has been conducted that resulted in an improved dynamic temperature model (Two-Zone Temperature and Solute Model) (Neilson 2006). This model has been developed to assist in more representative quantification of energy and mass exchange processes taking place within a river. This model incorporates in-situ solute and temperature measurements for model population and calibration. Essentially, the purpose of this model is to predict instream temperature based on various boundary condition and multiple parameters allowing for improved management decision.

As part of the data collection effort, thermal infrared (TIR) remotely sensed spatial temperature data was collected of a 28.3 kilometer reach of the Virgin River near Hurricane Utah on June 22, 2007. This airborne TIR imagery was collected using a remote sensing forward-looking infrared (FLIR) instrumentation (FLIR Systems P65 camera (North Billerica, MA, USA)), which was deployed by a fixed-wing single prop aircraft. In addition to the TIR imagery, 3-band imagery was collected simultaneously.

2. Objectives

The TIR and 3-band imagery collection effort was conducted based on the hypothesis that a minimum of two currently estimated physical parameters could be replaced with two measured parameters. These parameters are average area of river surface area associated with dead zone and average channel widths. Dead zone in this context is slower moving water, in relation to the velocity of the main channel flows, generally found near the banks of the river and behind debris. Currently, these two parameters are being estimated versus physically measured and only one value for each parameter is being used for the entire study reach. In other words, an average channel width of 20 meters and an average area of 30 percent of the channel is estimated to be dead zone is being used to depict these physical parameters of the entire 28.3 km reach, which is clearly not the most accurate representation of these two parameters. Therefore, to hypothetically improve model calibration and reduce the number of estimated parameters in the model is the objective of this project to use TIR and 3-band imagery to physically measure these parameters along the entire 28.3 km reach. Thus the replacement of these currently estimated physical parameters with multiple measured physical parameters will hypothetically improve model prediction accuracy, which will provide for improved water management decisions.

3. METHODOLOGY

3.1. Data Collection

As previously mentioned, TIR and 3-band imagery was collected via a single prop fixed wing aircraft. The resolution of this imagery was 0.7 meters. Along with the imagery, instream temperatures were collected simultaneously with in-situ temperature probes. This in-situ data was to aid in image correction.

3.2. Image Processing

The TIR imagery, as previously mentioned, was corrected for atmospheric interference using ground-truthing temperature data and MODTRAN. And a mosaic of the TIR and 3-band imagery was generated using ERDAS IMAGINE (Norcross, GA).

3.3. River Mask/River Polygon

Once the mosaics of the images were generated, a mask of water pixels was created using ERDAS IMAGINE (Norcross, GA) based on 3-band image pixel classifications. This mask of the water was in the form of a raster with which a polygon of the river was generated.

3.4. River Temperature Raster

The river polygon allowed the clipping of the TIR imagery to produce a raster of river temperatures. The temperature raster of the river is the keystone to this project. It allowed the calculation of dead zone areas and channel widths to be completed.

3.5. River Temperature Raster Corrections

The river temperature raster was then clipped into 3 segments based on knowledge of differing influences on instream temperatures. For example, the lower segment is influenced by cool water discharge from a warming pond used to warm a hypolymnatic dam release. Therefore, the release from the pond is cooler in relation to the upstream water temperatures. Thus, below the confluence of the pond release and the Virgin River water temperatures are cooler for roughly two kilometers. This was accomplished by generating 3 new shape files in ArcCatology and using the SNAP tool to ensure that the edge of each polygon lined up properly, thus allowing accurate clips of the raster of river temperature.

With the three segments of the river temperature raster completed, the next step was to determine whether or not the pixels of each raster were actually water pixels versus land pixels. This was accomplished with the aid of the 3-band imagery. Since the resolution of both the 3-band and TIR imagery is 0.7 meters, determining land from water is visually straightforward. Additionally, the use of the “identify tool” of ArcMap to determine large temperature gradients between pixels assisted in determining land from water. With this tool, an average maximum temperature was calculated (36^oC). The following conditional statement was then entered into the raster calculator which assigned the value 0 to temperatures greater then 36^oC.

Con([raster]>36,0,[raster])

This produced a raster of only temperature less then 36^oC. Furthermore, under the propertiesàsymbologyàclassificationàdata exclusion the exclusion statement; -1-0 was entered which excludes all pixels with the values between -1^oC and 0^oC. These steps allowed the correction of the river temperature raster to contain only water temperature pixels and no land pixels. This procedure was run on all three segments.

3.6. Dead Zone/Main Channel Raster Break

Now with three rasters of river temperature that essentially only contain water temperature pixels, the temperature break was ready to be calculated, meaning at what temperature to divide main channel pixels from dead zone pixels. The In-situ stream temperature data, mentioned previously, was used to determine the appropriate temperature break which would assign pixels above this certain temperature a color and pixels below with another color. This color scheme assists in visualizing the dead zone from main channel.

3.7. Dead Zone Area

The percentage of river for all three segments was then able to be calculated using the data exclusion feature of the classifications portion of the symbology tab under the properties table of each raster. First, the only pixels that were excluded were those of the temperature range -1-0^oC, which resulted in the total number of pixels for the respective river segment. The total number of pixels was used to calculate the total area of that segment. Then the pixels from -1^oC to the specified dead zone break temperature were excluded, which provide the number of pixels associated with dead zone temperatures. Thus, a total area associated with dead zone was calculated. With these two areas calculated for each river segment a percentage of the channel could be assigned as dead zone, which is one of the parameters of the model that was to be generated as a result of this project.

The calculation of the percentage of channel associated as dead zone is to be more precise and will be explained in more detail below.

3.8. Center Line/31m Point/River Sub Polygon Generation

Using the river polygon a center line was generated. This line is the center line of the river polygon, therefore it essentially follows the thalwaug of the river. This line was generated so as to be able to create a point file with a point every 31 meters down the center line. The purpose of the 31m points file is to generate 31m long sub polygons of the river polygon, thus allowing improved average channel width and percentage of channel associated with dead zone. Therefore, instead of an average channel width and dead zone area for three river segments, these values will be calculated for approximately 910 segments. It is hypothesized that this will improve the accuracy of both parameters; channel width and dead zone area, thus improving model calibration.

4. RESULTS/DISCUSSION

4.1. Image Processing

As previously mentioned, a mosaic of both the TIR and 3-band imagery was created using ERDAS Imagine (Norcross, GA) (Figure 1 and Figure 2, respectively). The resolution of both these mosaics is 0.7 meter, which allows for great visualization of water versus land. The high resolution was necessary for creating the mask of the river.

Figure 1. Mosaic of TIR imagery of 28.3km reach of the Virgin River.

Figure 2. Mosaic of 3-Band imagery of 28.3km reach of the Virgin River

4.2. River Mask/River Polygon

The results of the river mask can be seen in Figure 3. This is a close-up section of the river to depict the different classifications. This mask was generated by Dr Christopher Neale of USU using ERDAD IMAGINE (Norcross, GA).

Figure 3. The raster mask generated of the Virgin River showing a close-up view of a section of the Virgin River.

The resulting river polygon that was created with the mask can be seen in Figure 4. As seen in this figure there is a large body of water that is not part of the Virgin River. Therefore, to generate accurate dead zone and total river surface area this polygon needed to be edited. The result of the edit may be seen in Figure 5.

Figure 4. The river polygon of the Virgin River prior to edit. Note the large water body on east side of river.

Figure 5. The river polygon of the Virgin River post edit. Note the absences of the large water body on east side of river.

4.3. River Temperature Raster

The result of clipping the temperature raster with the river polygon can be seen in Figure 6. From close analysis of this raster it was clear that some of the pixels contained in the river temperature raster were actual land pixels versus water pixels. The completion of this step led to the next step.

Figure 6. River temperature raster of Virgin River.

4.4. River Temperature Raster Corrections

As a result of the river polygon not perfectly delineating the edge of water, the river temperature raster contained not only pixels that represented water temperatures but land temperatures also. This can be seen in Figure 7 through 9. From Figure 7 and 8, it is clear that the sand bars/islands within the red boxes are included in the river temperature raster as water which is not correct. It was necessary to remove any pixels that represented land temperature so as to not provide an erroneous river surface area calculation. The raster calculation resulted in a temperature raster that contained only pixels with temperature values between 1 and 36^oC, which proved more representative of the river proper (Figure 9).

Figure 7. Segment of Virgin River (3 band image).

Figure 8. Temperature Raster overlaid on 3 band image before raster calculation. Note the red boxes indicating regions with land temperature pixels.

Figure 9. River temperature raster post raster calculation. Note the islands within the red boxes that were previous represented by hot temperatures.

The total area prior and post raster calculation was calculated based on the total number of pixels and is contain in Table 1. With a minimum percent difference of 15 and a maximum of 25 it is obvious these corrections were necessary in order to provide representative river surface areas.

Table 1. Contains the total area calculations pre and post raster calculation.

Sub-Reach	Pre-Correction	Post-Correction	Difference	%
	Total Area (m^2)	Total Area (m^2)	(m^2)	Difference
Lower	210525	179301	31223	15
Middle	145115	113423	31692	22
Upper	207029	154386	52644	25

4.5. Dead Zone/Main Channel Raster Break

With the river temperature raster of each segment corrected the next step was to create a “dividing temperature” which would separate dead zone from main channel. Figure 10 is an example of this division between main channel and dead zone. The red pixels in this figure represent dead zone and yellow the main channel. It was assumed the dead zones near the banks would be found to be the slower moving waters, in relation to the main channel. This is not always the case but it was for the time this imagery was collected, which was the afternoon in June, thus the warmest time of day for

this region.

Figure 10. Main channel and dead zone are represented in this figure. Dead zone being the red pixels and main channel being yellow pixels.

4.6. Dead Zone/Main Channel Area

The calculated dead zone and main channel areas are contained in Table 2. As suspected, the area of dead zone for each sub-reach varies. These reaches have different temperature influences; hence the generations of sub reaches of the study reach. The upper reach is highly influenced by the presence of a hot spring that discharges into this sub-reach. The lower sub-reach is influence by an inflow from a warming pond. The purpose of the warming pond is to raise the water temperature of hypolymnotic dam release prior to re-entry to the Virgin River. The water temperature of the pond release is cooler then the upstream water temperature, therefore the instream temperatures below the pond release are cooler then the upstream temperature for approximately 2 kilometers downstream until water has had prolonged exposure time to solar radiation

Table 2. Contains the calculated main channel and dead zone areas as well as the percentage area of each zone.

Sub-Reach	DZ Area (m^2)	MC Area (m^2)	% DZ	% MC
Lower	104423	74878	27	73
Middle	45932	67491	40	60
Upper	59435	94951	38	62

4.7. Center Line/31m Point Generation

The results of the center line and 31m spaced points can be seen in Figure 11.

Figure 11. Shows the centerline and 31m points which were used to generate river sub polygons.

4.8. River Sub Polygons/Dead Zone Areas/Channel Widths

The current results of the river sub polygons can be seen in Figure 12. The generation of all the sub polygons is still in progress; therefore the average dead zone areas and average channels widths of these polygons have not yet been calculated. When completed a total of 910 sub polygons will be generated. Once the average dead zone areas and average channel widths are calculated for each sub polygon this data will be used to populate the model. These 910 values of dead zone area and average channel widths are hypothesized to be an immense improvement on the one average channel width and one average dead zone area for the entire 28.3 km reach.

Figure 12. Shows the centerline and 31m points which were used to generate river sub polygons.

5. CONCLUSION

From the TIR and 3-band imagery collected in June 2007, it is apparent that total river surface area dead zone and main channel areas can be calculated for the flow rate at which the imagery was recorded. Although these areas are not exact calculations due to the inability to remove all land temperature pixels from the river raster, also to include every water temperature pixel, they are an improvement on current parameter estimations. Further work is necessary to decrease the data processing and analysis time if this is to be repeated multiple times for various flow regimes. Perhaps this work should include some visual basic coding in conjunction with ArcGIS to automate the dead zone area and average channel width calculations for each of the 910 sub-polygons. This would significantly reduce the time requirements to make these calculations.

Additionally, the calculation of all 910 average channel widths will be a vast improvement on the one average channel width previously used for the entire reach. These values are still to be generated. Also, these values have yet to be incorporated into the model and the output of the model reviewed. This will validate all that has been done thus far. Recoding the model to be able to incorporate the data collected from this project was beyond the scope of this project and therefore was not done. It is proposed that it will be incorporated in the future.

Overall, this project has shown that with the proper analysis, dead zone and main channel areas may be calculated using TIR and 3-band imagery. Whether or not these calculations improve model calibration is yet to be determined.

References

Benson, B. B. and D. Krause (1980). "The concentration and isotopic frationation of gases dissiloved in freshwater in equilibrium with the atmosphere." Limnology and Oceanography 25(4): 662-670.

EPA. (2004). "Top 100 Impairments List." Retrieved #TOP_IMP, from http://oaspubepagov/waters/national_reptcontrol.

Holtby, L. B. (1988). "Effects of logging on stream temperatures in Carnation Creek, British Columbia, and associated impacts on the coho salmon (Oncorhynchus kisutch)." Canadian Journal of Fisheries and Aquatic Sciences 45: 502-515.

Neilson, B. (2006). Dynamic Stream Temperature Modeling: Understanding the Causes and Effects of Temperature Impairments and Uncertainty in Predictions. Logan, Utah State University.

Quigley, T. M. and S. J. Arbelbide (1997). An assessment of ecosystem components in the interior Columbia basin and portions of the Klamath and Great Basins. USDA Forest Service Pacific Northwest Research Station General Technical Report. 3.

Stoneman, C. L. and M. L. Jones (2000). "The influence of habitat features on the biomass and distribution of three species of southern Ontario stream salmonines. " Transactions of the American Fisheries Society 129: 639-657.

Wissmar, R. C., J. E. Smith, et al. (1994a). "A history of resource use and distrurbance in riverine basins of estern Oregon and Washington (early 1800s-1900s)." Northwest Sceince 68: 1-35.